version 3:
	- Fix Makefile for Mac OS.
	- The torch wrapper now takes inputs images where the pixel values go from 0 to 1 (it used to be from 0 to 255) to harmonize with the torch image library.
	- The option -f now returns the output of layer 18 (it used to be layer 16), so that the output should have size 4096x1x1 if the input has size 3x231x231.
	- Add python wrapper.
	- Add Mac OS binaries.

version 3.1:
	- Fix overfeat script so that the correct layer is produced for the large network.

version 3.2:
	- Fix bug in python wrapper that caused some images to be incorrectly loaded.
	- Better python wrapper sample script.

version 3.3:
	- Fix overfeat script that resized the images even when extracting the features.
	- Changed the default output layer of the overfeat script when using feature extraction.
	- Separate files from weights and add download_weights.py script to download the weights.

version 3.4:
	- Added binaries for linux 32 bits.
	- Fixed wrong output layer in overseat with option -f.

version 3.5:
	- Fix LD_LIBRARY_PATH in overfeat script.
	- Add git repository.

version 3.6:
	- Fix ppm images reading that didn't support comments in the ppm image.
	- Fix wrong paths in overfeat_batch.
	- Fix layer numbers in README.

version 4.0:
	- Released binaries for GPU (only for linux 64bits for now).
	- Removed dependency to torch by adding local copy of lib TH.
	- Replaced dynamic library libTH.so by static library libTH.a.
	- Fix bug in webcam script.
	- Fixed ambiguous dependency to openblas : now, linux uses openblas and macos uses Accelerate.
	- Mac OS should correctly use OpenMP now.